Mailing-List: contact dev-help@hive.apache.org; run by ezmlm
Precedence: bulk
Reply-To: dev@hive.apache.org
Date: Sat, 15 Mar 2014 22:32:43 +0000 (UTC)
From: "Lefty Leverenz (JIRA)" <jira@apache.org>
To: hive-dev@hadoop.apache.org
Message-ID: <JIRA.12622858.1354855000253.77029.1394922763026@arcas>
In-Reply-To: <JIRA.12622858.1354855000253@arcas>
References: <JIRA.12622858.1354855000253@arcas>
Subject: [jira] [Commented] (HIVE-3777) add a property in the partition to
 figure out if stats are accurate
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Transfer-Encoding: quoted-printable


    [ https://issues.apache.org/jira/browse/HIVE-3777?page=3Dcom.atlassian.=
jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=3D13936=
327#comment-13936327 ]=20

Lefty Leverenz commented on HIVE-3777:
--------------------------------------

Documented in the wiki's Configuration Properties, please review:

{quote}
hive.stats.reliable
Default Value: false
Added In: Hive 0.10.0 with HIVE-1653
New Behavior In:  Hive 0.13.0 with HIVE-3777

Whether queries will fail because statistics cannot be collected completely=
 accurately. If this is set to true, reading/writing from/into a partition =
or unpartitioned table may fail because the statistics could not be compute=
d accurately. If it is set to false, the operation will succeed.

In Hive 0.13.0 and later=EF=BB=BF, if hive.stats.reliable is false and stat=
istics could not be computed correctly, the operation can still succeed and=
 update the statistics but it sets a partition property "areStatsAccurate" =
to false. If the application needs accurate statistics, they can then be ob=
tained in the background.
{quote}

Questions:=20

# Does an unpartitioned table have the "areStatsAccurate" property too?
# Does the new behavior happen when hive.stats.reliable is false, not true?=
  (I ask because the jira description implies that this is a fix for the pr=
oblem of long-running queries failing when statistics aren't accurate, but =
as I understand it the query doesn't fail when hive.stats.reliable is false=
.  Perhaps I'm confused, so please make sure the wikidoc is correct.)

Quick ref:
* [Language Manual -- Configuration Properties:  hive.stats.reliable |https=
://cwiki.apache.org/confluence/display/Hive/Configuration+Properties#Config=
urationProperties-hive.stats.reliable]

> add a property in the partition to figure out if stats are accurate
> -------------------------------------------------------------------
>
>                 Key: HIVE-3777
>                 URL: https://issues.apache.org/jira/browse/HIVE-3777
>             Project: Hive
>          Issue Type: Improvement
>          Components: Query Processor
>    Affects Versions: 0.13.0
>            Reporter: Namit Jain
>            Assignee: Ashutosh Chauhan
>             Fix For: 0.13.0
>
>         Attachments: HIVE-3777.2.patch, HIVE-3777.2.patch, HIVE-3777.3.pa=
tch, HIVE-3777.4.patch, HIVE-3777.5.patch, HIVE-3777.patch
>
>
> Currently, stats task tries to update the statistics in the table/partiti=
on
> being updated after the table/partition is loaded. In case of a failure t=
o=20
> update these stats (due to the any reason), the operation either succeeds
> (writing inaccurate stats) or fails depending on whether hive.stats.relia=
ble
> is set to true. This can be bad for applications who do not always care a=
bout
> reliable stats, since the query may have taken a long time to execute and=
 then
> fail eventually.
> Another property should be added to the partition: areStatsAccurate. If h=
ive.stats.reliable is
> set to false, and stats could not be computed correctly, the operation wo=
uld
> still succeed, update the stats, but set areStatsAccurate to false.
> If the application cares about accurate stats, it can be obtained in the=
=20
> background.


--
This message was sent by Atlassian JIRA
(v6.2#6252)